Spatial discriminant analysis based on power-elliptic distributions and power transformations

Consider the classification of land-cover categories based on multispectral data. In this study, we transform the observed multivariate data by the Box-Cox transform for accelerating normality. We examine transformations for marginal distributions and for joint distributions. Then, the ordinary classification methods are applied to the transformed data. The method is applied to actual Landsat 5 data, and it is shown that our methods improve classification accuracies